Submission 218
Better Hatch Network
Submitter group: Civil society.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Better Hatch Network argues AI adoption in animal agriculture should be encouraged where it reduces animal suffering, using in-ovo sexing as its example, and regulated where it risks increasing it. It addresses welfare floors, risk-proportionate classification, standards-setting independence, auditable data and consumer defaults in AI shopping agents.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Says that Australia has no legislative equivalent to Germany's OKT model for in-ovo sexing, and that adoption so far has been industry-driven.
Set against Germany's OKT model, Australia still has no legislative equivalent for the technology already in commercial use here. Adoption to date has been industry-driven, without a legal requirement or a certification standard comparable to Germany's mark.
Argues that compliance obligations should depend on what an AI system does, not the industry it sits in.
a narrow, sensor-based classifier used to sex an egg carries a fundamentally different risk profile from a generative or autonomous decision-making system, whatever industry either operates in. Compliance obligations should track that difference in what a system does, not which industry it happens to sit in.
Argues that, judged only on productivity and profitability, AI adoption in the sector risks repeating past welfare harms at greater speed.
Judged only against productivity and profitability, AI adoption in this sector has every reason to repeat the scale of harm battery cages and selective breeding created, at a pace faster than either did.
Argues that AI oversight in animal agriculture needs records that are both technically trustworthy and legally accessible.
A reliable record and an auditable one are separate requirements, and AI oversight in this sector needs both. The record should be technically trustworthy: interoperable, tamper-evident logs that cannot be selectively edited after the fact. And it should be legally reachable: available to a regulator or court by default, not withheld as commercial-in-confidence.
Argues that the default values in AI shopping agents are presently set by whoever builds the agent, without disclosure to or oversight by consumers.
Whether an agent treats such a value as a default a consumer must opt out of, or as a request the consumer must actively make, will materially affect outcomes at scale. That decision is presently made by whoever builds the agent, without disclosure to, or oversight by, the consumer it affects.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Include welfare or sentient-impact assessment alongside productivity, safety and privacy criteria in agriculture AI frameworks, as a floor that never constrains technology reducing welfare harm faster or further.
That any AI standards or governance framework the Committee recommends that covers agriculture include a welfare or broader sentient-impact assessment as a floor, alongside existing productivity, safety and privacy criteria, not as a ceiling: it should never constrain an AI-enabled technology that reduces welfare harm faster or further than the floor requires.
Preferentially weight, or condition, government AI-adoption funding in animal agriculture on technologies reducing rather than risking entrenchment of long-term welfare harm.
That government funding, grants or incentive programs supporting AI adoption in animal agriculture preferentially weight, or condition support on, technologies that reduce rather than risk entrenching long-term welfare harm.
Have government look to scale a workable existing model, either a legislated requirement paired with a consumer label as in Germany or an industry-led certification standard like HFAC's, rather than design a parallel one from first principles.
That government look to scale a workable existing model, whether a legislated requirement paired with a consumer label, as in Germany, or an industry-led certification standard, as HFAC's is, rather than design a parallel one from first principles;
Extend Australia's mandatory egg-labelling standard to cover the welfare category that in-ovo sexing creates, described as a direct, low-cost step.
extending Australia's mandatory egg-labelling standard to cover the welfare category in-ovo sexing is creating would be a direct, low-cost step in that direction.
Base any future risk-based classification of AI systems on what a system does, not its industry, distinguishing narrow task-specific applications from general-purpose or higher-risk systems so obligations stay proportionate.
That any future risk-based classification of AI systems under Australian law or standards consider what a system does, not the industry it operates in: distinguishing narrow, task-specific applications, such as sensor-based classification used in agriculture, from general-purpose or higher-risk decision-making systems, including future systems within agriculture itself, so that compliance obligations remain proportionate to actual risk.
Ensure standards for AI systems monitoring or managing animal welfare in agriculture are not developed or validated solely by industry-affiliated bodies.
That standards for AI systems used to monitor or manage animal welfare in agriculture not be developed or validated solely by industry-affiliated bodies
Keep data from AI welfare-monitoring systems in interoperable, tamper-evident logs with an independent access pathway for regulators, not commercial-in-confidence by default.
That data generated by AI welfare-monitoring systems in agriculture be kept in interoperable, tamper-evident logs, with a clear independent access pathway for regulators, rather than defaulting to commercial-in-confidence treatment.
Consider shorter review or sunset periods for AI standards than government’s general default, provided review capacity is resourced to match.
Given how quickly the underlying technology is moving, standards in this space may also warrant shorter review or sunset periods than government's general default, provided review capacity is resourced to match.
Support the Productivity Commission's recommended independent National Animal Welfare Commission, with AI-related standards on review cycles short enough to catch goal drift and track adoption.
That the Committee support the Productivity Commission's recommendation for an independent National Animal Welfare Commission, with review cycles for any AI-related standards kept short enough to catch goal drift in general-purpose systems and track the pace of adoption.
Require AI commercial agents operating in Australia to disclose the default preference settings, beyond data handling and privacy, that they apply on a consumer's behalf.
That AI commercial agents operating in Australia be required to disclose the default preference settings, beyond data handling and privacy, that they apply on a consumer's behalf.
Require AI commercial agents to set preset defaults to the harm-minimising option unless and until a consumer actively chooses otherwise.
That AI commercial agents be required to set preset defaults to the harm-minimising option, unless and until a consumer actively chooses otherwise.
Consider whether a fiduciary-style obligation, like that proposed for US custodial user agents, has a place in any future Australian framework, with attention to consumer values rather than data use alone.
That the Committee consider whether a fiduciary-style obligation, of the kind proposed for the US's custodial user agents, has a place in any future Australian framework, with explicit attention to consumer values rather than data use alone.